alterlab-paper-reviewer
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration attempts were identified in the skill instructions or agent definitions. The skill correctly utilizes its scoped tools for academic analysis and integration with the AlterLab suite.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted academic manuscripts which represent a potential surface for indirect prompt injection. \n1. Ingestion points: Manuscript text enters the system via the $ARGUMENTS variable in the review-paper command and is subsequently read by analysis agents. \n2. Boundary markers: The instructions for the agents do not explicitly implement boundary markers or instructions to disregard embedded directives within the processed manuscripts. \n3. Capability inventory: The skill environment has access to Read, Write, Edit, Bash, WebFetch, and WebSearch tools, which are used to verify citations and evaluate research quality. \n4. Sanitization: No explicit sanitization or input filtering is performed on the manuscript content before it is processed by the agent personas.
Audit Metadata